Yul: introduce YulNameRepository

libyul/CMakeLists.txt

@@ -43,6 +43,8 @@ add_library(yul
 	YulControlFlowGraphExporter.cpp
 	YulName.h
 	YulString.h
+	YulName.cpp
+	YulName.h
 	backends/evm/AbstractAssembly.h
 	backends/evm/AsmCodeGen.cpp
 	backends/evm/AsmCodeGen.h

libyul/YulName.cpp

@@ -0,0 +1,162 @@
+/*
+	This file is part of solidity.
+
+	solidity is free software: you can redistribute it and/or modify
+	it under the terms of the GNU General Public License as published by
+	the Free Software Foundation, either version 3 of the License, or
+	(at your option) any later version.
+
+	solidity is distributed in the hope that it will be useful,
+	but WITHOUT ANY WARRANTY; without even the implied warranty of
+	MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+	GNU General Public License for more details.
+
+	You should have received a copy of the GNU General Public License
+	along with solidity.  If not, see <http://www.gnu.org/licenses/>.
+*/
+// SPDX-License-Identifier: GPL-3.0
+
+#include <libyul/YulName.h>
+
+#include <libyul/Exceptions.h>
+
+#include <fmt/compile.h>
+
+namespace solidity::yul
+{
+
+YulNameRepository::YulNameRepository()
+{
+	auto const emptyName = defineName("");
+	yulAssert(emptyName == YulNameRepository::emptyName());
+}
+
+std::optional<std::string_view> YulNameRepository::labelOf(YulName _name) const
+{
+	if (!isDerivedName(_name))
+	{
+		// if the parent is directly a defined label, we take that one
+		auto const labelIndex = std::get<0>(m_names[static_cast<size_t>(_name)]);
+		yulAssert(labelIndex <= std::numeric_limits<size_t>::max());
+		return m_definedLabels.at(static_cast<size_t>(labelIndex));
+	}
+	return std::nullopt;
+}
+
+std::string_view YulNameRepository::requiredLabelOf(YulName const& _name) const
+{
+	auto const label = labelOf(_name);
+	yulAssert(label.has_value(), "YulName currently has no defined label in the YulNameRepository.");
+	return label.value();
+}
+
+YulNameRepository::YulName YulNameRepository::baseNameOf(YulName _name) const
+{
+	while (isDerivedName(_name))
+		_name = std::get<0>(m_names[static_cast<size_t>(_name)]);
+	return _name;
+}
+
+std::string_view YulNameRepository::baseLabelOf(YulName const& _name) const
+{
+	auto const labelIndex = std::get<0>(m_names[static_cast<size_t>(baseNameOf(_name))]);
+	yulAssert(labelIndex <= std::numeric_limits<size_t>::max());
+	return m_definedLabels[static_cast<size_t>(labelIndex)];
+}
+
+std::optional<YulNameRepository::YulName> YulNameRepository::nameOfLabel(std::string_view const _label) const
+{
+	auto const it = std::find(m_definedLabels.begin(), m_definedLabels.end(), _label);
+	if (it != m_definedLabels.end())
+	{
+		YulName labelName{static_cast<YulName>(std::distance(m_definedLabels.begin(), it))};
+		// mostly it'll be iota
+		if (!isDerivedName(labelName) && std::get<0>(m_names[static_cast<size_t>(labelName)]) == labelName)
+			return std::get<0>(m_names[static_cast<size_t>(labelName)]);
+		// if not iota, we have to search
+		auto itName = std::find(m_names.rbegin(), m_names.rend(), std::make_tuple(labelName, YulNameState::DEFINED));
+		if (itName != m_names.rend())
+			return std::get<0>(*itName);
+	}
+	return std::nullopt;
+}
+
+YulNameRepository::YulName YulNameRepository::defineName(std::string_view const _label)
+{
+	if (auto const name = nameOfLabel(_label); name.has_value())
+		return *name;
+
+	m_definedLabels.emplace_back(_label);
+	m_names.emplace_back(m_definedLabels.size() - 1, YulNameState::DEFINED);
+	return static_cast<YulName>(m_names.size() - 1);
+}
+
+YulNameRepository::YulName YulNameRepository::deriveName(YulName const& _name)
+{
+	m_names.emplace_back(baseNameOf(_name), YulNameState::DERIVED);
+	return static_cast<YulName>(m_names.size() - 1);
+}
+
+bool YulNameRepository::isDerivedName(YulName const& _name) const
+{
+	yulAssert(_name <= std::numeric_limits<size_t>::max());
+	return std::get<1>(m_names.at(static_cast<size_t>(_name))) == YulNameState::DERIVED;
+}
+
+void YulNameRepository::generateLabels(std::set<YulName> const& _usedNames, std::set<std::string> const& _illegal)
+{
+	std::set<std::string> used;
+	std::set<YulName> toDerive;
+	for (auto const name: _usedNames)
+	{
+		if (!isDerivedName(name))
+		{
+			auto const label = labelOf(name);
+			yulAssert(label.has_value());
+			auto const [it, emplaced] = used.emplace(*label);
+			if (!emplaced || _illegal.count(*it) > 0)
+				// there's been a clash ,e.g., by calling generate labels twice;
+				// let's remove this name and derive it instead
+				toDerive.insert(name);
+		}
+		else
+			yulAssert(isDerivedName(name) || _illegal.count(std::string(*labelOf(name))) == 0);
+	}
+
+	std::vector<std::tuple<std::string, YulName>> generated;
+	auto namesIt = _usedNames.begin();
+	for (size_t nameValue = 1; nameValue < m_names.size(); ++nameValue)
+	{
+		auto name = static_cast<YulName>(nameValue);
+		if (namesIt != _usedNames.end() && name == *namesIt)
+		{
+			if (isDerivedName(name) || toDerive.find(name) != toDerive.end())
+			{
+				std::string const baseLabel(baseLabelOf(name));
+				std::string label (baseLabel);
+				size_t bump = 1;
+				while (used.count(label) > 0 || _illegal.count(label) > 0)
+					label = fmt::format(FMT_COMPILE("{}_{}"), baseLabel, bump++);
+				if (auto const existingDefinedName = nameOfLabel(label); existingDefinedName.has_value())
+				{
+					std::get<0>(m_names[static_cast<size_t>(name)]) = *existingDefinedName;
+					std::get<1>(m_names[static_cast<size_t>(name)]) = YulNameState::DEFINED;
+				}
+				else
+					generated.emplace_back(label, name);
+				used.insert(label);
+			}
+			++namesIt;
+		}
+	}
+
+	for (auto const& [label, name] : generated)
+	{
+		yulAssert(_illegal.count(label) == 0);
+		m_definedLabels.emplace_back(label);
+		std::get<0>(m_names[static_cast<size_t>(name)]) = static_cast<YulName>(m_definedLabels.size() - 1);
+		std::get<1>(m_names[static_cast<size_t>(name)]) = YulNameState::DEFINED;
+	}
+}
+
+}

libyul/YulName.h

@@ -18,9 +18,87 @@
 
 #pragma once
 
+#include <libyul/ControlFlowSideEffects.h>
+#include <libyul/SideEffects.h>
 #include <libyul/YulString.h>
 
+#include <fmt/format.h>
+
+#include <unordered_map>
+#include <memory>
+#include <vector>
+#include <string>
+#include <optional>
+#include <functional>
+#include <numeric>
+#include <map>
+
 namespace solidity::yul
 {
+
+enum class LiteralKind;
+struct Block;
+
+/**
+ * Yul Name repository.
+ *
+ * A database of numerical identifiers for Yul nodes in an AST (`YulName`s). Identifiers can be
+ *
+ *  - 'defined', i.e., they are equipped with a string label, which can be retrieved by a call to `labelOf(yulName)`;
+ *  - 'derived', i.e., they don't possess a string label but have a parent yul name which is either also derived or
+ *    defined. All dependency chains of derived labels terminate in a defined label.
+ *
+ * Such derived identifiers can be introduced, e.g., during certain optimization steps. If the AST (or segments thereof)
+ * should be printed and/or the string label of identifiers should be retrieved which are still in derived state,
+ * a call to `generateLabels(identifiers)` changes the status of all derived labels to defined and generates
+ * unique labels for all identifiers provided to the method. The generated labels are based on their parents.
+ */
+class YulNameRepository
+{
+public:
+	using YulName = std::uint64_t;
+	YulNameRepository();
+	/// Defines a new name based on a label. If the label already was defined, it returns the corresponding YulName
+	/// instead of a new one (`defineName("xyz") == defineName("xyz")`).
+	YulName defineName(std::string_view _label);
+
+	/// Defines a new name based on a parent name. When generating labels, the generated label will be based on the
+	/// parent's (`deriveName(id) != deriveName(id)`).
+	YulName deriveName(YulName const& _baseName);
+
+	/// The empty name.
+	static constexpr YulName emptyName() { return 0; }
+
+	/// Yields the label of a yul name. The name must have been added via `defineName`, a label must have been
+	/// generated with `generateLabels`, or it is a builtin.
+	std::optional<std::string_view> labelOf(YulName _name) const;
+
+	/// If it can be assumed that the label was already generated, this function will yield it (or fail with an
+	/// assertion error).
+	std::string_view requiredLabelOf(YulName const& _name) const;
+
+	/// Yields the label of the base name of the provided name. Opposed to `labelOf`, this must always exist.
+	std::string_view baseLabelOf(YulName const& _name) const;
+
+	/// Whether a name is considered derived, i.e., has no label but a parent name.
+	bool isDerivedName(YulName const& _name) const;
+
+	/// Tries to find the label in the defined names and returns the corresponding name.
+	std::optional<YulName> nameOfLabel(std::string_view label) const;
+
+	/// Generates labels for derived names over the set of _usedNames, respecting a set of _illegal labels.
+	/// This will change the state of all derived names in _usedNames to "not derived" with a label associated to them.
+	void generateLabels(std::set<YulName> const& _usedNames, std::set<std::string> const& _illegal = {});
+
+private:
+	enum class YulNameState	{ DERIVED, DEFINED };
+
+	/// Yields the name that the provided name was based on - or the name itself, if the name was directly "defined".
+	YulName baseNameOf(YulName _name) const;
+
+	std::vector<std::string> m_definedLabels{};
+	std::vector<std::tuple<YulName, YulNameState>> m_names{};
+};
 using YulName = YulString;
+
 }

test/CMakeLists.txt

@@ -164,6 +164,7 @@ set(libyul_sources
     libyul/SyntaxTest.cpp
     libyul/YulInterpreterTest.cpp
     libyul/YulInterpreterTest.h
+    libyul/YulName.cpp
     libyul/YulOptimizerTest.cpp
     libyul/YulOptimizerTest.h
     libyul/YulOptimizerTestCommon.cpp

test/libyul/YulName.cpp

@@ -0,0 +1,86 @@
+/*
+	This file is part of solidity.
+
+	solidity is free software: you can redistribute it and/or modify
+	it under the terms of the GNU General Public License as published by
+	the Free Software Foundation, either version 3 of the License, or
+	(at your option) any later version.
+
+	solidity is distributed in the hope that it will be useful,
+	but WITHOUT ANY WARRANTY; without even the implied warranty of
+	MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+	GNU General Public License for more details.
+
+	You should have received a copy of the GNU General Public License
+	along with solidity.  If not, see <http://www.gnu.org/licenses/>.
+*/
+// SPDX-License-Identifier: GPL-3.0
+
+#include <boost/test/unit_test.hpp>
+
+#include <libyul/Dialect.h>
+#include <libyul/YulName.h>
+#include <libyul/backends/evm/EVMDialect.h>
+
+using namespace solidity;
+using namespace solidity::yul;
+
+namespace solidity::yul::test
+{
+
+BOOST_AUTO_TEST_SUITE(YulName)
+
+BOOST_AUTO_TEST_CASE(repository_generate_labels_for_derived_types)
+{
+	Dialect dialect{};
+	YulNameRepository nameRepository;
+	auto const test1 = nameRepository.defineName("test1");
+	BOOST_CHECK(!nameRepository.isDerivedName(test1));
+	auto const test1_1 = nameRepository.deriveName(test1);
+	BOOST_CHECK(nameRepository.isDerivedName(test1_1));
+	auto const test1_2 = nameRepository.deriveName(test1);
+	BOOST_CHECK(nameRepository.isDerivedName(test1_2));
+	auto const test1_2_1 = nameRepository.deriveName(test1_2);
+	BOOST_CHECK(nameRepository.isDerivedName(test1_2_1));
+	auto const test2 = nameRepository.defineName("test2_1");
+	auto const test2_1 = nameRepository.deriveName(test2);
+	auto const test3 = nameRepository.defineName("test3");
+	auto const test3_1 = nameRepository.deriveName(test3);
+	BOOST_CHECK(test1 != test1_1);
+	BOOST_CHECK(test1 < test1_1);
+	nameRepository.generateLabels({test1, test1_1, test1_2, test1_2_1, test2_1, test3, test3_1}, {"test1"});
+
+	// marking test1 as invalid means that all labels get bumped up by one
+	BOOST_CHECK(nameRepository.labelOf(test1) == "test1_1");
+	BOOST_CHECK(nameRepository.labelOf(test1_1) == "test1_2");
+	BOOST_CHECK(nameRepository.labelOf(test1_2) == "test1_3");
+	BOOST_CHECK(nameRepository.labelOf(test1_2_1) == "test1_4");
+
+	BOOST_CHECK(nameRepository.labelOf(test2) == "test2_1");
+	// the label of test2 is reused as it's not in the used names when generating labels
+	BOOST_CHECK(nameRepository.labelOf(test2_1) == "test2_1");
+
+	BOOST_CHECK(nameRepository.labelOf(test3) == "test3");
+	BOOST_CHECK(nameRepository.labelOf(test3_1) == "test3_1");
+
+	// derive a name from the (now labelled) test2_1 name
+	auto const test2_1_1 = nameRepository.deriveName(test2_1);
+	// but we have a conflict with an already defined/labelled name, expectation is that we get test2_1_2
+	auto const conflict = nameRepository.defineName("test2_1_1");
+	nameRepository.generateLabels({test1, test1_1, test1_2, test1_2_1, test2_1, test2_1_1, test3, test3_1, conflict});
+	// test2_1 is in the list, so produce a new name
+	BOOST_CHECK(nameRepository.labelOf(test2_1_1) == "test2_1_2");
+	BOOST_CHECK(nameRepository.labelOf(conflict) == "test2_1_1");
+
+	nameRepository.generateLabels({test2, test2_1, test2_1_1, conflict});
+	BOOST_CHECK(nameRepository.labelOf(test2) == "test2_1");
+	// this label gets reassigned, as test2_1 is back in the game
+	BOOST_CHECK(nameRepository.labelOf(test2_1) == "test2_1_3");
+	BOOST_CHECK(nameRepository.labelOf(test2_1_1) == "test2_1_2");
+	BOOST_CHECK(nameRepository.labelOf(conflict) == "test2_1_1");
+
+}
+
+BOOST_AUTO_TEST_SUITE_END()
+
+}